israeli-survivor-benefits-navigator
Fail
Audited by Gen Agent Trust Hub on Aug 21, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions direct the AI agent to execute a local helper script (
scripts/estimate_survivor_allowance.py) using command-line arguments derived directly from user input. Because the instructions do not specify any validation or sanitization requirements for these inputs (such as 'age', 'orphans', or 'wage'), a malicious user could provide inputs designed to break out of the command string and execute arbitrary shell commands. - Evidence: 'Run scripts/estimate_survivor_allowance.py --age --orphans --seniority-years to produce the estimate' in SKILL.md.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a broad attack surface for indirect prompt injection due to its requirement to process diverse and untrusted user inputs (family status, financial data) which are then used to drive calculations and external research.
- Ingestion points: User-provided data regarding family relationships, ages of survivors/orphans, cause of death, and deceased's wages (SKILL.md, Step 1).
- Boundary markers: Absent. The skill provides no instructions for the agent to use delimiters or to disregard instructions potentially embedded within user inputs.
- Capability inventory: The skill has the capability to execute a local Python script and perform network lookups via the agent's tools (SKILL.md, Step 3 and Step 6).
- Sanitization: Absent. There are no directives to sanitize or escape user input before it is interpolated into script arguments or the agent's reasoning context.
Recommendations
- AI detected serious security threats
Audit Metadata